A Hybrid Kalman-Weighted Sliding Mode Observer for Sensorless Torque Estimation of Robotic Manipulators

Abstract

Accurate sensorless external force estimation is crucial for physical human-robot interaction. To address the challenge that existing momentum-based sliding mode observers face in simultaneously achieving fast dynamic response and effective chattering suppression, this paper proposes a Kalman-weighted adaptive second-order sliding mode observer (HKW-SOSMO). This method utilizes the discrete Riccati equation to compute the joint posterior covariance in real time, employing it as a dynamic weight to modulate the sliding mode switching gain. The gain is adaptively amplified in regions with sudden friction changes, while it decreases in steady-state regions as the covariance contracts. Based on Lyapunov theory, this paper proves the finite-time convergence of this variable-gain system. Simulation results demonstrate that the proposed method effectively resolves the trade-off between dynamic response and chattering, thereby significantly enhancing estimation accuracy.